Lesson Plan - ML24ECSC306

School of Computer Science and Engineering
FMTH0301/Rev.5.3
Course Plan
Semester: 5 Year: 2024-25

Course Title: Machine Learning and Deep Learning Course Code: 24ECSC306
Total Contact hrs.: 6 hrs Duration of ESA: 2 Hrs.
ISA Marks: 50 ISA Marks: 50
Lesson Plan Author: Dr. Uday Kulkarni, Sunil V Gurlahosur Date: 16-09-2024
Checked By: Dr. Meena S M Date: 23-09-2024
Prerequisites: Basics of Python programming, Exploratory Data Analysis, Linear Algebra, Calculus,
and Statistical models.
Course Overview:
The course is designed to enhance critical thinking, creativity, and innovation, enabling students to not
only keep up with current technological trends but also contribute to solving future domain-specific
challenges. The course covers key concepts like supervised and unsupervised learning, ensemble
learning, and seq2seq methodologies, providing a comprehensive understanding of modern AI
techniques. Through course projects, students develop a responsible approach to AI, ensuring their
work positively impacts society while preparing them for practical engineering applications in this rapidly
evolving field, often leading to valuable research publications.
Course Outcomes (COs):

At the end of the course the student should be able to:
1. Explain supervised and unsupervised machine learning algorithms.

2. Apply various ensemble learning methods to improve the performance of the machine learning
models.
3. Implement Seq2Seq models for sequence-related multimedia tasks.
4. Employ appropriate machine learning algorithms for a given real-world application.
5. Create a comprehensive report of course project and publish paper at technical conference.
1
Course Articulation Matrix: Mapping of Course Outcomes (COs) with Program Outcomes
(POs)
Course Title: Machine Learning and Deep Learning Semester: 5
Course Code: 24ECSC306 Year: 2024-25
Course Outcomes (COs) / Program 1 2 3 4 5 6 7 8 9 10 11 12 13 14

Outcomes (POs)
1. Explain supervised and H M
unsupervised machine learning
algorithms.
2. Apply various ensemble learning H H
methods to improve the
performance of the machine
learning models.
3. Implement Seq2Seq models for H H
sequence-related multimedia
tasks.
4. Employ appropriate machine H H
learning algorithms for a given
real-world application.
5. Create a comprehensive report of H H H
course project and publish paper
at technical conference
Degree of compliance L: Low M: Medium H: High
2
Course Content
Course Code: 24ECSC306 Course Title: Machine Learning and Deep Learning
L-T-P : 2-0-2 Credits: 4 Contact Hrs.: 6 Hrs.
ISA Marks: 50 ESA Marks: 50 Total Marks: 100
Teaching Hrs: 30 Lab Hrs: 56 ESA Theory Exam Duration: 2 hrs
Content Hrs
Unit - 1
Chapter 1: Introduction and Regression: Fundamentals of ML, linear, ridge, lasso, 4 hrs.
elastic-net regression, evaluation.
Chapter 2: Classification: Linear discriminant analysis, logistic regression, support 5 hrs.
vector machines, decision tree, extra trees, Bayesian networks, evaluation.
Chapter 3: Ensemble learning: Bagging, boosting, stacking, random forest, resampling 6 hrs.
methods.
Unit - 2
Chapter 4: Neural Networks: Perceptron, gradient descent, optimization algorithms, 5 hrs.
backpropagation, hyper parameters, regularization.
Chapter 5: Deep Neural Networks: convolutional neural networks, various CNN 6 hrs.
architectures, model selection and evaluation, bias-variance.
Chapter 6: Seq2Seq models: Recurrent neural networks, long short-term memory, auto 4 hrs.
encoders.
Text Books (List of books as mentioned in the approved syllabus)
1. Tom Mitchell., Machine Learning, Mc Graw Hill, McGraw-Hill Science, 3rd edition.
2. Ian Goodfellow and Yoshua Bengio and Aaron Courville: Deep Learning, MIT Press, 2016.
References
1. Aurelian Gerona, Hands-On Machine Learning with Scikit-Learn and Tensor Flow, Concepts,
Tools, and Techniques to Build Intelligent Systems, Publisher: O'Reilly Media, July 2016.
2. Luca Pietro Giovanni Antiga, Thomas Viehmann, Eli Stevens, Deep Learning with PyTorch
Manning Publications, 2020.
Evaluation Scheme (ISA)
SL. Section Evaluation Chapter Numbers Conducted Weightage

No. Scheme for marks in Marks
3
i. ISA -I 1, 2, 3 30 25 Marks
ii. ISA – II 4, 5, 6 30
iii. Lab Evaluation ISA 1, 2, 3, 4, 5, 6 20 10 Marks
iv. Course Project 1, 2, 3, 4, 5, 6 30 15 Marks
Evaluation Scheme (ESA)
SL. Section Evaluation Chapter Numbers Conducted Weightage

No. Scheme for marks in Marks
Theory Exam 1, 2, 3, 4, 5, 6 60
i. 25 Marks
(ESA)
Lab Exam (ESA) ESA Report/Paper & Poster 10
ii. Demo/modification 10 25 Marks
Viva 5
Minimum 20 marks in ISA for eligibility.
4
In-Semester Assessment Scheme
Course Unitization for Minor Exams and End Semester Assessment
No. of No. of No. of No. of

Topics / Chapters Teaching Questions Questions Questions Questions
Credits in Minor in Minor in Activity in ESA
Exam-1 Exam-2
Unit I
Introduction and Regression 4 1 - - 1
Classification 5 1 - - 1
Ensemble learning 6 1 - - 1
Unit II
Neural Networks 6 - 1 - 1
Deep Neural Networks 5 - 1 - 1
Seq2Seq models 4 - 1 - 1
Note
1. Each Question carries15 marks and may consists of sub-questions.
2. Mixing of sub-questions from different chapters within a unit (only for Unit I and Unit II) is allowed in
Minor I, II and ESA
3. Answer 4 full questions of 15 marks each (two full questions from Unit I, II) out of 6 questions in
ESA.
5
Course Assessment Plan
Course Title: Machine Learning and Deep Learning Code: 24ECSC306
Assessment Methods
Lab
Weightage in activity
Course outcomes (COs)
assessment ISA1 ISA2 and Lab Theory
course
project
1. Explain supervised and
unsupervised machine ✓ ✓ ✓ ✓
40%
learning algorithms.
2. Apply various ensemble
learning methods to
improve the performance of ✓
✓ ✓ ✓
the machine learning 10%
models.
3. Implement Seq2Seq models
for sequence related ✓ ✓ ✓ ✓
10%
multimedia tasks.
4. Employ appropriate
machine learning algorithms
✓ ✓
for a given real-world 20%
application.
5. Create a comprehensive
report of course project and
publish paper at technical ✓ ✓
20%
conference.
Weightage 25% 50% 25%
6
Chapter wise Plan
Course Code and Title: 24ECSC306 / Machine Learning and Deep Learning
Chapter Number and Title: 1. Introduction to Regression Planned Hours: 4 hrs
Learning Outcomes:-
At the end of the topic, the student should be able to:
Topic Learning Outcomes COs BL CA Code

1. Distinguish between different kinds of machine learning
systems. 1 L2 1.3
2. Explain supervised, unsupervised and reinforcement learning 1 L2 1.3
3. Identify situations where linear regression is appropriate and 1 L3 1.3
apply linear regression models.
4. Apply different Regularization methods to the model. 1 L3 2.1
5. Choose a proper cost function for a given problem 1 L3 2.3
Lesson Schedule
Class No. - Portion covered per hour
1.Introduction to Machine Learning and its applications
2. Introduction to Linear Regression
3. Gradient Descent for Linear Regression
4. Regularization - Ridge and Lasso and elastic Net
Review Questions
Sl.No. - Questions TLOs BL PI Code
1. What is machine learning? What is a hypothesis? What are the 1 L2 1.3.1
three main components of the machine learning process?
2. Define features, observations, and hypotheses. What are the 3 L2 1.3.1
various data formats of a dataset? How does data format affect
machine learning tasks? Explain with a suitable example.
3. Design a system that predicts the salary of a person based on 3 L3 1.3.1
his experience using the machine learning approach. What will
be experience E to learn task T to perform P for the system?
4. Refer to the data given in the table below, if in the linear 5 L3 2.3.1
regression model θ0 = 0, find the cost using the cost function for θ1 = 0,
0.5 and 1. Which is the best value for θ1?
population of district (in lakhs) number of hospitals

1 2
7
2 5
3 10
4 14
5. Explain the importance of regularization in machine learning, 4 L2 2.1.3
describe how it works, and discuss common regularization
techniques.
6. Explain the role of the following factors in reaching global minima 3 L3 1.3.1
with a gradient descent algorithm for linear regression.
a. Epochs
b. Learning rate
c. Parameters
d. Bias and Variance
7. Explain overfitting in linear regression with examples. How do 4 L2 2.1.3
you overcome overfitting?
8. Describe the bias-variance tradeoff in machine learning. Why is 4 L2 2.1.3
it important, and how does it affect model performance?
9. Compare L1 and L2 regularization techniques in terms of their 4 L3 2.1.3
effects on model complexity and feature selection. How do they
differ in their impact on the magnitude of weight values?
10. Explain the difference between supervised, unsupervised and 2 L2 1.3.1
reinforcement learning techniques with suitable examples.
8
Chapter wise Plan
Chapter Number and Title: 2. Classification Planned Hours:5 hrs
Learning Outcomes:-
At the end of the topic, the student should be able to:
Topic Learning Outcomes COs BL PI Code

1. Differentiate between PCA and LDA. 1 L2 2.1
2. Use logistic regression to make predictions and classifications. 1 L3 2.3
3. Choose a proper cost function for a given classification problem. 1 L3 2.3
4. Use regularization to solve the overfitting problem. 1 L3 2.3
5. Apply various classification techniques. 1 L3 2.3
Lesson Schedule
Class No. - Portion covered per hour
1. Linear Discriminant Analysis
2. Introduction to Logistic Regression and Gradient Descent
3. Support Vector Machines
4. Decision trees
5. Extra trees and Bayesian networks
Review Questions
Sl.No. - Questions TLOs BL PI
Code
1. Explain Linear Discriminant Analysis and how it performs dimensionality 1 L2 2.1.3
reduction.
2. Why it is necessary to estimate the accuracy of hypothesis. Explain with an 4 L3 2.3.1
example for logistic regression using regularization.
3. The logistic regression model does not calculate the cost using the sum of 3 L3 2.3.1
square of errors. Do you agree with the statement? If yes, explain why and how
is cost calculated in logistic regression? If no, then justify.
4. Apply SVM algorithm for the data points in Table-1 and find dimension of 5 L3 2.3.1
hyperplane to classify them.
X Y Label
4 2 -1
4 -2 -1
6 1 -1
9
6 -1 -1
8 0 1
10 -1 1
12 2 1
12 -2 1
Table-1
5. Use an appropriate decision tree algorithm on a dataset given below and 5 L3 2.3.1
determine whether to play football or not.
Outlook Temperature Humidity Wind Played
football(yes/no)
Sunny Hot High Weak No
Sunny Hot High Strong No
Overcast Hot High Weak Yes
Rain Mild High Weak Yes
Rain Cool Normal Weak Yes
Rain Cool Normal Strong No
Overcast Cool Normal Strong Yes
Sunny Mild High Weak No
Sunny Cool Normal Weak Yes
Rain Mild Normal Weak Yes
Sunny Mild Normal Strong Yes
Overcast Mild High Strong Yes
Overcast Hot Normal Weak Yes
Rain Mild High Strong No
6. What are Bayesian Networks and How does the Kernel Function works. 5 L2 2.3.1
7. Consider a dataset with features (X) and a label (Y): 5 L3 2.3.1
X = [[5, 3], [4, 7], [6, 8], [3, 9], [8, 2]]
Y = [0, 1, 1, 0, 1]
Classify a new data point [7, 5] and make a final prediction based on majority
voting by building a decision tress with maximum depth 2.
10
Chapter Number and Title: 3: Ensemble Learning Planned Hours: 6 hrs.
Learning Outcomes: -
At the end of the topic the student should be able to:

1. Apply the concepts of bagging to identify appropriate classes on given 2 L3 2.1
dataset.
2. Implement the Random Forest technique for the given problem. 2 L3 2.3
3. Apply various resampling methods to improve the performance of the 2 L3 2.3
classification model.
4. Build multi-class classifiers using ensemble learning techniques. 2 L3 2.1
5. Develop the AdaBoost technique with proper parameters to solve the 2 L3 2.3
classification problem.
6. Apply stacking technique for a given dataset. 2 L3 2.3
Lesson Schedule
Class No. - Portion covered per hour / per Class
1. Introduction to Ensemble learning
2. Bagging technique
3. Boosting and stacking technique
4. Random Forest technique
5. AdaBoost technique
6. Resampling method
Review Questions
1. Explain how does Ensemble learning help in Incremental Learning? 4 L2 2.1.3
2. Discuss with suitable examples and reasons when we should use and not 4 L3 2.1.3
use the ensemble Learning improves the performance of machine learning
models?
3. Explain how Bagging and Boosting technique help to improve the 5 L2 2.3.1
performance of the model?
4. Consider a regression problem with 'n’ predictions on test data by ‘n’ 1 L3 2.1.3
different models (M1, M2, …. Mn) respectively. Explain the methods which
can be used to combine the predictions of these models?
5. Can we ensemble multiple models of same machine learning algorithm? 3 L3 2.3.1
Justify you answer with suitable reason and example.
11
6. Discuss how accuracy is achieved when using different resampling 3 L2 2.3.1

methods?
8. List and explain the different components of a stacked ensemble model. 6 L2 2.3.1
What is the role of base model and meta model in them.
9. In practice, Random Forest can be susceptible to overfitting. Explain the 2 L3 2.3.1
potential reasons behind overfitting in Random Forest and suggest
strategies or techniques to mitigate this issue.
10. Compare and contrast the AdaBoost algorithm with other ensemble 4 L3 2.3.1
methods, such as Random Forest and Gradient Boosting. Highlight the
key differences in their boosting strategies, model diversity, and overall
predictive performance. Provide insights into when AdaBoost might
outperform these alternatives.
11. Perform Bagging on the following dataset for 5 sample iteration. 4 L3 2.1.3
X 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Y 1 1 1 -1 -1 -1 -1 1 1 1
12
Chapter Number and Title: 4: Neural Network Planned Hours: 5 hrs.
Learning Outcomes: -

1. Explain the core principles and mathematics behind perceptron. 1 L3 2.1
2. Implement gradient descent for neural networks. 1 L3 2.4
3. Apply forward and back propagation for model training. 1 L3 2.4
4. Build multi-class classifiers using neural networks. 1 L3 2.3
5. Build efficient neural network by choosing appropriate hyper parameters. 1 L3 2.3
6. Apply the different regularization methods to build robust models. 1 L3 2.3
Lesson Schedule
1. Introduction to perceptron learning, Model representation
2. Gradient checking
3. Back propagation algorithm
4. Hyper parameter Tuning, Multiclass Classification
5. Applications of Neural Network, Regularization
Review Questions:

1. “Neural networks are much better for a complex nonlinear hypothesis 1 L3 2.1.3
even when feature space is huge “. Do you agree with the above statement?
Justify with suitable example.
2. What is multilevel classification? Explain the same for neural network with 4 L2 2.3.1
appropriate example.
3. Discuss how accuracy is achieved when varying the number of hidden 5 L2 2.3.1
layers and number of neurons at each layer in a fully connected neural
network.
4. Consider the following Neural Network with α=0.0001 η=0.25, desired 2 L3 2.4.1
output=1 and sigmoid activation function.
• Perform forward pass and calculate the error.
• Calculate the updated weights for w5 and w6 using back-
propagation.
13
5. In the context of perceptrons, discuss the concept of the "XOR 1 L3 2.1.3

problem." How does it illustrate the limitations of single-layer
perceptrons, and what architectural changes are required to solve it?
6. Describe the tradeoff between the learning rate and convergence 5 L3 2.3.1
speed in gradient descent. How do you choose an appropriate learning
rate for a given optimization problem?
7. Discuss the concept of regularization methods and its importance in 6 L2 2.3.1
optimizing machine learning models. What are some common
regularization methods, and how do you choose them?
8. What is the purpose of a loss and activation function, and how is it 3 L3 2.4.1
involved in both forward and backpropagation? How does it measure
and help to improve the performance of a neural network during
training?
14
Chapter Number and Title: 5: Deep Neural Network Planned Hours: 6 hrs.
Learning Outcomes:

1. Explain how CNN excel at feature extraction and their ability to 1 L3 2.1
automatically learn various computer vision tasks.
2. Explain various operations, architectural innovations, design principles 1 L3 2.4
of different CNN architectures.
3. Apply cross validating and hyper parameter tuning for optimizing model 1 L3 2.3
performance.
4. Design efficient CNN architectures to overcome overfitting and 1 L3 2.4
underfitting through regularization and early stopping.
Lesson Schedule
1. Introduction to Deep Neural Network (DNN) and implementation
2. Implementation of DNN
3. Study of various DNN architecture
4. Model Selection and Evaluation
5. Model Selection and Evaluation
6. Bias-Variance tradeoff
Review Questions:

1. Describe in brief how Face Recognition is done using Convolutional 1 L2 2.1.3
Neural Networks.
2. List and explain the steps involved in training CNN from scratch. 2 L2 2.4.1
Describe the effects of learning rate and number of hidden layers and
their types on efficiency of neural network
3. Explain the concept of dropout regularization in deep neural networks. 4 L3 2.4.1
How does dropout prevent overfitting, and what is the ideal dropout rate
for a given network architecture?
4. Explain the concept of data augmentation in the context of CNNs, and 2 L3 2.4.1
provide examples of augmentation techniques. How can data
augmentation improve model generalization?
5. Discuss the advantages and disadvantages of using skip connections 2 L3 2.4.1
(residual connections) in deep neural networks, with a focus on the
impact on model training and convergence.
15
6. Consider a CNN architecture with 10,000 images with a batch size of 1 L3 2.1.3
64, with each input as 128 * 128, undergoes a convolution operation
with a kernel size of 3 * 3 and a stride of 2. Later max pooling layer is
applied with a pool size of 2 * 2 and stride of 2. Answer the following:
• How many iterations (or batches) will be required to complete
one training epoch?
• How many output feature maps (channels) will you get, and
what will be the dimensions of each output feature map for both
the Convolution and max pooling layer?
7. In a binary classification problem, your model predicts 30 true positives, 1 L3 2.1.3
10 true negatives, 5 false positives, and 15 false negatives. Calculate
the precision, recall, F1-score, and accuracy of the model.
8. Describe the challenges and potential solutions for handling class 3 L3 2.3.1
imbalance in a multi-class image classification problem when using
CNNs.
9. (i) What are the problems of VGGNet architecture? Explain how 4 L3 2.4.1
ResNet overcome these problems?
(ii) What are fully convolutional layers? In what situations are fully
convolutional layers better than fully connected layers and vice-versa?
10. Describe the following optimization techniques and list the advantages 3 L3 2.3.1
and disadvantages.
(i) SGD (ii) Gradient Descent with Momentum (iii) RMSProp (iv) ADAM
16
Chapter Number and Title: 6. Seq2Seq Models Planned Hours: 4 hrs.
Learning Outcomes:

1. Distinguish between CNN and Seq2seq models. 3 L2 2.1
2. Explain RNN and LSTM. 3 L2 2.1
3. Explore various applications of Seq2seq models. 3 L3 2.1
4. Understand efficient data representations using autoencoders. 3 L3 2.1
Lesson Schedule
1. Introduction to Seq2seq models, RNN
2. LSTM
3. Applications of Seq2seq models
4. Autoencoders
Review Questions
1. What is recurrent neural network? Why recurrent neural networks? How 2 L2 2.1.3
does RNN work?
2. What are various applications of recurrent neural networks? Discuss how 3 L3 2.1.3
image captioning works?
3. Explain how recurrent neural networks are different from convolution 1 L2 2.1.3
neural networks?
5. What are autoencoders? How data is represented using autoencoders? 4 L2 2.1.3
What are the various applications of autoencoders?
6. Consider the linear autoencoder over the real numbers. Show that all the 2 L3 2.1.3
information about the data is contained in the mean and covariance matrix
of the data. Show that the standard least square error function is a quadratic
function (parabola) in each individual weight, if all the other weights are
assumed to be constant.
7. Discuss when RNNs fail? How does long short-term memory (LSTM) 2 L2 2.1.3
work to overcome these failures?
17
8. Explain the concept of denoising autoencoders. How do they work, and 4 L3 2.1.3
what are the advantages of using denoising autoencoders in
applications like image or speech denoising.
Model Question Paper for In-Semester Examination (ISA-I)

Duration: 60 mins Max. Marks: 30
Note: Answer any two full questions 2. Each full question carries equal marks.
Q.No Questions Marks CO BL PO PI
Code
Calculate the cost value of linear regression for the following
dataset.
X = [1,2,3,4,5]
1.a Y = [3,6,7,11,15] 10 1 L3 1 2.3.1
Consider value of slope of hypothesis (Ɵ1) = 7
Constant/ intercept (Ɵ0) = 5
Epoch=3, Learning rate (α) = 0.5
List and explain how the Random Forest algorithm can be
b. classification, regression, and anomaly detection with 5 2 L2 2 2.3.1
examples.
Consider a dataset with features (X) and a label (Y):
X = [[5, 3], [4, 7], [6, 8], [3, 9], [8, 2]]
Y = [0, 1, 1, 0, 1]
2.a 10 1 L3 2 2.3.1
Classify a new data point [7, 5] and make a final prediction
based on majority voting by building a decision tress with
maximum depth 2.
Consider a student placement scenario.
i. Develop a hypothesis to predict whether a student will
b. be placed or not by considering appropriate features. 5 1 L2 1 2.3.1
Discuss cases of under-fitting, and over-fitting and how to
overcome these issues.
Perform Bagging on the following dataset.
Use 5 samples for prediction.
3.a 10 2 L3 2 2.1.3
X 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Y 1 1 1 -1 -1 -1 -1 1 1 1
Why gradient descent algorithm is required in logistic
regression. List the difference between stochastic gradient
b. 10 1 L2 1 2.3.1
descent and batch gradient descent with suitable cost
function.
18
Model Question Paper for In-Semester Examination (ISA-II)

Note: 1. Answer any two full questions 2. Each full question carries equal marks.
Q.No Questions PI
Marks CO BL PO
Code
1.a (i) You are given the function f(x,y) = x2 + xy + y2 and
are trying to find a local minimum using gradient
descent. You randomly start with x = 1.3 and y = 5.4.
Perform the first step of gradient descent with learning
rate α = 0.01. Show the resulting values for x and y as
well as all of your calculations.
10 1 L2 2 2.1.3
(ii) If we train a neural network for 1,000 epochs (one
training example at a time), does it make a difference
whether we present all training examples in turn for
1000 times or whether we first present the first training
example 1000 times, then the second training example
for 1000 times, and so on? Why?
b. Discuss when RNNs fail? How does long short-term

memory (LSTM) work to overcome these failures? 5 3 L3 2 2.4.1
2.a (i) What are the problems of VGGNet architecture?

Explain how ResNet overcome these problems?
(ii) What are fully convolutional layers? In what 10 1 L3 2
2.3.1
situations are fully convolutional layers better than fully
connected layers and vice-versa?
b. Describe the tradeoff between the learning rate and
convergence speed in gradient descent. How do you
choose an appropriate learning rate for a given 5 3 L3 2 2.1.3
optimization problem?
3.a Describe how the various gats in the Long Short-Term

Memory (LSTM) network help to solve the vanishing
10 2.3.1
gradient issue and make it possible to handle long- 3 L3 2
range dependencies more effectively by overcoming
drawbacks in Recurrent Neural Network (RNN).
b. How does the choice of activation function (e.g., ReLU,
sigmoid, tanh) as a hyperparameter impact the 1 2.4.1
5 L3 2
performance of a neural network? What factors should
be considered when selecting an activation function?
19
Model Question Paper for End Semester Examination (ESA)

Note: 1. Answer any two full questions from Unit -I and Unit -II. 2. Each full question carries equal marks.
UNIT – I
Q.No Questions PI
Marks CO BL PO
Code
1.a Calculate the cost and parameter values of linear
regression for the following dataset using batch
gradient descent algorithm.
X= [5, 2, 1, 7, 6]
10 1 L3 1 2.3.1
Y= [9, 5, 3, 11, 10]
Consider the value of the slope (Ɵ1) = 0, Constant/
intercept (Ɵ0) = 0, Epoch=2, and Learning rate (α) =
0.5.
b. Explore how Bagging, Boosting, and Stacking
individually contribute to improving model
performance. Provide examples illustrating how each 5 2 L2 2 2.3.1
technique addresses different aspects, such as bias,
variance, and generalization.
2.a Apply the SVM algorithm for the data points in Table 1
and find the dimension of the hyperplane to classify
them.
Table-1
X Y Label
1 -1 -1
1 1 -1 10 1 L3 2 2.3.1
2 1 -1
3 0 -1
5 -1 1
5 1 1
6 -2 1
7 2 1
b. Analyze the phenomena of overfitting and underfitting

in the context of a Linear Regression model.
5 1 L2 1 2.3.1
Elaborate on various strategies to overcome these
issues, incorporating real-time examples to illustrate
the implementation of such techniques.
3.a Perform Bagging on the following dataset. Use 5
samples for prediction.
10
X 0. 0. 0. 0. 0. 0. 0. 0. 0. 1 2 L3 2 2.1.3
1 2 3 4 5 6 7 8 9
Y 1 1 1 -1 -1 -1 -1 1 1 1
b. The logistic regression model does not calculate the
10 1 L2 1 2.3.1
cost using the sum of square of errors. Do you agree
20
with the statement? If yes, explain why and how is cost

calculated in logistic regression? If no, then justify.
Unit – II
4.a Consider a Multi-Layer Perceptron (MLP) architecture

in Figure 1, comprising an input layer with three
neurons, a hidden layer with two neurons utilizing the
sigmoid activation function, and an output layer with
two neurons employing the sigmoid activation. Given
the following set of weights and biases, compute the
output values resulting from the forward pass through
the network and calculate the error assuming the target
labels y1 and y2 as [0.01, 0.99] with learning rate(α) =
0.5.
10 1 L3 2 2.1.3
Figure: 1 Multilayer Perceptron

b. Describe the architecture and working principles of an
autoencoder in solving the problem of feature 5 3 L2 2 2.1.3
extraction, image reconstruction, and text
summarization.
5.a Explain the architecture of a typical Convolutional
Neural Network (CNN), focusing on the layers involved
in handling complex visual data and their contributions
to creating an efficient CNN. In a CNN architecture,
consider the convolutional layer processing input
image of size 128x128x3 using 512 filters of
dimensions 3x3. Calculate the following values with the
stride set at one and no padding employed during the 10 L3 2
1
convolution operation. 2.3.1
1. The total number of trainable parameters
present in the convolutional layer.
2. The number of multiplications performed
during the convolution operation for a single
input image passing through this
convolutional layer.
b. Using MLPs, design the networks to replicate two input 5 1 L2 2 2.1.3

AND, OR logic gates.
6.a Describe how the various gates in the Long Short-Term
10 3 L3 2 2.1.3
Memory (LSTM) network help to solve the vanishing
gradient issue and make it possible to handle long-
21
range dependencies more effectively by overcoming

drawbacks in Recurrent Neural Network (RNN).
b. List and explain the steps involved in training CNN from
scratch. Describe the effects of learning rate and 1 2.4.1
5 L2 2
number of hidden layers and their types on efficiency
of neural network.
22

Lesson Plan - ML24ECSC306

Uploaded by

Copyright:

Available Formats

Lesson Plan - ML24ECSC306

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Lesson Plan - ML24ECSC306

Uploaded by

Copyright:

Available Formats

School of Computer Science and Engineering

Semester: 5 Year: 2024-25

Course Outcomes (COs):

1. Explain supervised and unsupervised machine learning algorithms.

Course Outcomes (COs) / Program 1 2 3 4 5 6 7 8 9 10 11 12 13 14

Degree of compliance L: Low M: Medium H: High

Text Books (List of books as mentioned in the approved syllabus)

Evaluation Scheme (ISA)

SL. Section Evaluation Chapter Numbers Conducted Weightage

Evaluation Scheme (ESA)

SL. Section Evaluation Chapter Numbers Conducted Weightage

Minimum 20 marks in ISA for eligibility.

In-Semester Assessment Scheme

Course Unitization for Minor Exams and End Semester Assessment

No. of No. of No. of No. of

Course Assessment Plan

Course Title: Machine Learning and Deep Learning Code: 24ECSC306

Chapter wise Plan

Topic Learning Outcomes COs BL CA Code

population of district (in lakhs) number of hospitals

Chapter wise Plan

Topic Learning Outcomes COs BL PI Code

Topic Learning Outcomes COs BL CA Code

6. Discuss how accuracy is achieved when using different resampling 3 L2 2.3.1

X 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Topic Learning Outcomes COs BL CA Code

Sl.No. - Questions TLOs BL PI Code

5. In the context of perceptrons, discuss the concept of the "XOR 1 L3 2.1.3

Topic Learning Outcomes COs BL CA Code

Sl.No. - Questions TLOs BL PI Code

Chapter Number and Title: 6. Seq2Seq Models Planned Hours: 4 hrs.

Topic Learning Outcomes COs BL CA Code

Model Question Paper for In-Semester Examination (ISA-I)

Model Question Paper for In-Semester Examination (ISA-II)

b. Discuss when RNNs fail? How does long short-term

2.a (i) What are the problems of VGGNet architecture?

3.a Describe how the various gats in the Long Short-Term

Model Question Paper for End Semester Examination (ESA)

b. Analyze the phenomena of overfitting and underfitting

with the statement? If yes, explain why and how is cost

4.a Consider a Multi-Layer Perceptron (MLP) architecture

Figure: 1 Multilayer Perceptron

b. Using MLPs, design the networks to replicate two input 5 1 L2 2 2.1.3

range dependencies more effectively by overcoming

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.